基于块标记的田间叶片损伤区域分割方法  

Algorithm for damaged leaf segmentation based on block mark

作　　者：张水发[1,2] 王开义[1,2] 祖琴[3] 黄姗[1,2] 潘守慧[1,2] 王志彬[1,2] 李明远[4] 

机构地区：[1]北京农业信息技术研究中心,北京100097 [2]国家农业信息化工程技术研究中心,北京100097 [3]国家粮食局科学研究院,北京100037 [4]北京市农林科学院植保环保研究所,北京100097

出　　处：《农业工程学报》2014年第21期218-225,共8页Transactions of the Chinese Society of Agricultural Engineering

基　　金：农业部公益性行业科研专项项目支持(201203026)

摘　　要：为了在田间开放环境中有效分割叶片损伤区域,该文结合Canny算子良好的边缘提取能力和叶片局部颜色变化相对较小的特征,提出基于块标记的叶片损伤区域分割方法,用于评价叶片损伤程度。使用Android系统手机在晴天大田开放环境中采集木耳菜、西红柿、黄瓜、茄子、桃、彩椒和蛾眉豆7种常见农作物叶片图像,在阴天采集丝瓜、葫芦、甜瓜、茄子和黄瓜5种叶片图像,然后进行分割。该分割算法在晴天和阴天总体的平均正确分类率为97.5%,平均错误分类率为0.3%,并且有较好的目标一致性和边缘清晰度。应用系统对叶片损伤程度的评价结果与手工分割比较,在晴天和阴天采集图像上的平均误差分别为2.340%和1.475%,可较好地应用于晴天和阴天环境。该方法可探索应用于田间植物叶片损伤程度评价。Damaged leaf is one of the important factors leading to crop loss. Damaged leaf segmentation provides an important basis for diseased leaf detection, and for proper preventive measures to be taken. Advances in technology have made it possible for a computer with image processing techniques to segment the diseased leaf in an image of a green plant and evaluate the severity of the infestation. The research objects based on image segmentation and processing are the leaves damaged by pests or nutrient deficiency. The procedure of image segmentation algorithm was developed in C＋＋ that targets a diseased green leaf including the normal leaf and diseased regions. In current researches, algorithms based on thresholding or clustering are widely used. Despite of the simplicity and efficiency, the performances of these methods are not satisfactory due to the grayscale overlapping among background, plant leaves and damaged leaves in field environment. In consideration of the stability edge feature of images and the gray value consistency of leaves, a novel method was proposed to segment the damaged leaves in field environment by combining Canny edge detection and block mark, which is robust with respect to the changes in illumination and noises, and efficient to evaluate the damage degree of the leaves. Image processing was used to transform the image to gray scale, extract the Canny edge, perform Canny edge clustering, remove noise, detect the external rectangle, extract connected components which are 4-connected, classify regions, and finally segment the diseased regions of the green leaf. The block mark based algorithm was introduced to segment the damage leaf. The experiments were conducted on Malabar spinach, tomato, cucumber, eggplant, peach, pepper, dolichos lablab images captured on sunny day, and towel gourd, calabash, melon, eggplant and cucumber images captured on cloudy day. （1） The classification accuracy of the Malabar spinach on a sunny day was 98.8%; and 95.4%, 98.5%, 98.4%, 98.8%, 99.1%, 99.5% for to

关 键 词：计算机视觉 图像分割 病害控制 块标记 营养元素缺乏 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]